Accurate Assembly of Circular RNAs with TERRACE

Circular RNA (circRNA) is a class of RNA molecules that forms a closed loop with its 5' and 3' ends covalently bonded. Due to this specific structure circRNAs are more stable than linear RNAs, admit distinct biological properties and functions, and have been proven to be promising biomarkers. Circular RNAs were severely overlooked previously owing to the biases in the RNA-seq protocols and in the detection algorithms, but recently gained tremendous attentions in both aspects. However, most existing methods for assembling circRNAs heavily rely on the annotated transcriptomes, and hence exhibit unsatisfactory accuracy when a high-quality transcriptome is unavailable. Here we present TERRACE, a new algorithm for full-length assembly of circRNAs from paired-end total RNA-seq data. TERRACE uses the splice graph as the underlying data structure to organize the splicing and coverage information. We transform the problem of assembling circRNAs into finding two paths that "bridge" the three fragments in the splice graph induced by back-spliced reads. To solve this formulation, we adopted a definition for optimal bridging paths and a dynamic programming algorithm to calculate such paths, an approach that was proven useful for assembling linear RNAs. TERRACE features an efficient algorithm to detect back-spliced reads that are missed by RNA-seq aligners, contributing to its much improved sensitivity. It also incorporates a new machine-learning approach that is trained to assign a confidence score to each assembled circRNA, which is shown superior to using abundance for scoring. TERRACE is compared with leading circRNA detection methods on both simulations and biological datasets. Our method consistently outperforms by a large margin in sensitivity while maintaining better or comparable precision. In particular, when the annotations are not provided, TERRACE can assemble 123%-412% more correct circRNAs than state-of-the-art methods on human tissues. TERRACE presents a major leap on assembling full-length circRNAs from RNA-seq data, and we expect it to be widely used in the downstream research on circRNAs.